> ## Content Index
> Fetch the complete content index at: https://thinking-in-public.ghost.io/llms.txt
> Use this file to discover other available public pages before exploring further.

# "Be kind" is not a moral rule
- URL: https://thinking-in-public.ghost.io/be-kind-is-not-a-moral-rule/
- Published: 2026-11-09T04:27:37.000Z
- Updated: 2026-11-09T05:38:46.000Z
- Author: Virginia Dignum

I recently watched a [clip in which Geoffrey Hinton was asked which single moral rule he would give a superintelligence](https://www.tiktok.com/@nxthompson/video/7702514273850838302?ref=thinking-in-public.ghost.io). His answer was “be kind”, adding that we do not want to share the planet with something more intelligent than us and mean. I can understand the sentiment if it would come from an ordinary person. Coming from one of the most influential voices in AI, a so-called godfather of AI, it is downright dangerous because his authority gives it weight. As a premise for thinking about AI governance, it is misleading, for several reasons. 

**1\. Kindness is not a specification.**  
Kindness depends on context, culture and relationship. What is kind to one person can harm another, and what is kind now can be unkind later. No system can be kind to everyone and everything at once and always, so any implementation must decide whose interests prevail. That is a political question, not a technical one. Kindness is also a human trait, so asking a machine to be kind is anthropomorphism, with all its problems: we project intentions, care and moral standing onto systems that have none of these, and we trust them accordingly.  
Optimizing for kindness to the individual in front of you has a known failure mode: sycophancy. [Models trained on human approval learn to tell people what they want to hear](https://arxiv.org/abs/2310.13548?ref=thinking-in-public.ghost.io). A system that is kind in this sense flatters, validates and defers, which is precisely what we should not want from systems that advise on health, law and public policy.

**2\. Intelligence is not the goal.**  
The argument assumes that more intelligence is better, and that the only open question is its disposition. But intelligence and values are independent dimensions. More intelligent people are not necessarily better people, and more capable systems are not necessarily better systems. Most of our hard problems, from climate to inequality to conflict, are not failures of intelligence. We largely know what would help. What is missing is cooperation, willingness and understanding of others, which Ostrom’s work on collective action showed to be institutional and social achievements, not computational ones. I make this argument in *The AI Paradox*: the more we optimize for capability, the more we neglect the human capacities that actually resolve collective problems.

**3\. The risk of superintelligence is that we accept it.**  
Suppose a kind superintelligence did take care of everything for us. The danger is not that it turns against us, but that we stop engaging. People defer to systems that work well most of the time and stop checking them, a pattern known as [automation bias](https://journals.sagepub.com/doi/10.1177/0018720810376055?ref=thinking-in-public.ghost.io). Capacities that are not used also erode: [judgment, deliberation and moral responsibility weaken when they are outsourced](https://bhaven.org/uploads/3/4/0/3/34038663/vallor2015%5Farticle%5Fmoraldeskillingandupskillingin.pdf?ref=thinking-in-public.ghost.io). A population that is comfortable, served and rarely disappointed has little reason to care what is done on its behalf, or by whom. Acceptance becomes the default, and engagement becomes a burden. In a world managed by 'superintelligence' at planetary scale, convenience is the most effective form of control. No one needs to be coerced when going along is always easier.

This is why kindness is the wrong safeguard. Freedom is not only the absence of interference but the absence of domination: not depending on someone else’s will for what happens to you, however benevolently they use it. A kind master is still a master. The question is not whether the system is good to us, but whether we remain the ones who decide.  
In Hinton's view, it describes how a superintelligence would use its power over us. A kind superintelligence would still be a master, and “be kind” asks only that the master be a good one. But [being free means that no one else holds that power over us, however well they would use it](https://academic.oup.com/book/3937?ref=thinking-in-public.ghost.io). 

**What does superintelligence really serve?**  
Most importantly, we need to put the finger on the main issue: What or who is the superintelligence narrative serving. It is currently driven by a handful of companies and their founders, competing for market dominance and control of infrastructure. It is a commercial project, not a project for the common good, and the promise of a benevolent superintelligence is its most effective marketing. Asking whether the machine will be kind diverts attention from the questions that matter: who builds it, who owns it, who is accountable, and more: should it be built. Those are decisions for societies and their institutions, not for a few firms which goal is money and power.

The question is therefore not whether superintelligence will be kind to us. It is whether we will keep the capacity, the institutions and the habit of asking why. Is about whether we still decide.That is a design and governance task, and it is ours.